Frontier Lab Researcher: AI Safety Proposals Over-Focus on RSI, Miss Real-World Details

_aidan_clark_ · x · 2026-09-09

A thread on AI safety governance: a frontier AI lab researcher responds to safety policy proposals, arguing they are thoughtful but written by people far from the day-to-day details of frontier labs in a harmful way—abstracting over key difficulties, over-focusing on recursive self-improvement (RSI), and underfocusing on the concrete realities of the current AI landscape.

Related event: Ex-OpenAI researcher Aidan Clark questions whether AI is moving too fast(7 posts)→

Original post →

More from AGI Musings

AGI Musings channel →